Goodness of fit tests for the pseudo-Poisson distribution
share
Bivariate count models having one marginal and the other conditionals being of the Poissons form are called pseudo-Poisson distributions. Such models have simple exible dependence structures, possess fast computation algorithms and generate a sufficiently large number of parametric families. It has been strongly argued that the pseudo-Poisson model will be the first choice to consider in modelling bivariate over-dispersed data with positive correlation and having one of the marginal equi-dispersed. Yet, before we start fitting, it is necessary to test whether the given data is compatible with the assumed pseudo-Poisson model. Hence, in the present note we derive and propose a few goodness-of-fit tests for the bivariate pseudo-Poisson distribution. Also we emphasize two tests, a lesser known test based on the supremes of the absolute difference between the estimated probability generating function and its empirical counterpart. A new test has been proposed based on the difference between the estimated bivariate Fisher dispersion index and its empirical indices. However, we also consider the potential of applying the bivariate tests that depend on the generating function (like the Kocherlakota and Kocherlakota and Mu noz and Gamero tests) and the univariate goodness-of-fit tests (like the Chi-square test) to the pseudo-Poisson data. However, for each of the tests considered we analyse finite, large and asymptotic properties. Nevertheless, we compare the power (bivariate classical Poisson and Com-Max bivariate Poisson as alternatives) of each of the tests suggested and also include examples of application to real-life data. In a nutshell we are developing an R package which includes a test for compatibility of the data with the bivariate pseudo-Poisson model.
READ FULL TEXT
